RGB LEDs for color enlargers

[koraks]

256 Hz, and 50% is showing a slow rise/fall.

To be brutally honest, if that's an accurate measurement of light intensity, this is an exceptionally poor PWM performance. I'd be seriously concerned about this.

fuzz around the line -- that's 100 kHz ripple that I'd like to remove.

Do you mean the small-amplitude 'heavy' fuzz (2.5mV amplitude if I read correctly), or the much lighter fuzz that's apparently not there most/some of the time and only sporadically at some points of the period (the 15-20mV amplitude fuzz)? The latter looks like an oscillation; I'd be tempted to scope out the actual current through the LEDs to see if it's there, or if you're picking up a spurious effect in your phototransistor detector circuit.

I think the phototransistor is useful for determining thermal degradation effects, but perhaps less so for measuring PWM performance. To that end, I'd directly measure LED current instead.

If you’re using pwm dimming, a capacitor would cause trouble on low levels at a minimum I think

It would further degrade PWM slope performance, indeed. As such it would affect all duty cycles, with very low ones being erased away and very high ones all glued together to 100%. Depending on component choice / dimensioning, of course. I don't use any caps across the driver outputs for this reason. I do use a much higher PWM frequency; I think I'm running the LEDs at 10kHz now. This used to be 1.5kHz in my previous setup. I'd have to go back to my notes on the buck frequency I chose for the LED drivers, but off the top of my head I set this at 330kHz.

 

[albada]

To be brutally honest, if that's an accurate measurement of light intensity, this is an exceptionally poor PWM performance. I'd be seriously concerned about this.

Do you mean the small-amplitude 'heavy' fuzz (2.5mV amplitude if I read correctly), or the much lighter fuzz that's apparently not there most/some of the time and only sporadically at some points of the period (the 15-20mV amplitude fuzz)? The latter looks like an oscillation; I'd be tempted to scope out the actual current through the LEDs to see if it's there, or if you're picking up a spurious effect in your phototransistor detector circuit.

I think the phototransistor is useful for determining thermal degradation effects, but perhaps less so for measuring PWM performance. To that end, I'd directly measure LED current instead.

It would further degrade PWM slope performance, indeed. As such it would affect all duty cycles, with very low ones being erased away and very high ones all glued together to 100%. Depending on component choice / dimensioning, of course. I don't use any caps across the driver outputs for this reason. I do use a much higher PWM frequency; I think I'm running the LEDs at 10kHz now. This used to be 1.5kHz in my previous setup. I'd have to go back to my notes on the buck frequency I chose for the LED drivers, but off the top of my head I set this at 330kHz.

I fixed a mistake: The power-supply for the phototransistor circuit had some noise, and adding a cap eliminated that.
Second, my cheap scope has much A2D noise at 5 mv/div, and that's the uniform band around all lines. The taller spikes are ripple from the Mean Well driver.

But here's an interesting discovery that is probably the cause of the bad PWM performance koraks observed: The Mean Well driver performs some RC smoothing of the PWM input, which the datasheet didn't mention. In fact, the datasheet has a graph showing the output PWM is identical to the input. Here's a trace of 1/16th duty-cycle at 256 Hz:

At 1/64th duty-cycle, the output is a nearly flat line.
This smoothing behavior is good to know about, because the reciprocity characteristics of paper might differ from pulsed vs continuous light.

 

[koraks]

This smoothing behavior is good to know about, because the reciprocity characteristics of paper might differ from pulsed vs continuous light.

I think you're mixing up two things here, but correct me if I'm wrong.
One concern is high-intensity, short duration reciprocity failure. The kind of effect you might run into when using flash photography with a very short thyristor-truncated flash pulse of let's say 1/10,000 second. While film generally doesn't suffer from this down to about 1/1000 second (1 millisecond), I don't know about paper. My experience so far suggests it's not much of a concern, however. To the best of my knowledge, no test results have ever been released to the general public reporting about rapidly repeated short pulsed exposure like we have in PWM!
The other concern is how your PWM circuitry handles very short duty cycles. In the example you've shown, you encounter the problem that very short pulses actually don't make it to the LEDs at all! That has nothing to do with reciprocity effects, but it's just a fundamental shortcoming of the Meanwell LED driver you're using. It implies that if you somehow end up using very short duty cycles with that setup, you won't expose the paper at all, or only much less so than you anticipated. This is the issue I brought to the fore in my previous post and that would concern me.

Mind you, at some point you pretty much always run into very short duty cycles blending to a flat line. However, I had this happen at something like 0.5% duty cycles at 10kHz (beyond this everything seemed linear enough), not at 1.5% duty cycles at 256Hz. The latter might actually become problematic for split grade printing where you try to mix in relatively small amounts of blue with green to create the lower grades around 1 or so. You could solve this to an extent by using a non-linear relationship between grade and blue channel PWM duty cycle, but if the effect is bad enough, you'll effectively lose a part of the contrast range between grade 00 and grade 1.5 or so.

PS: I just wrote something about PWM resolution, although not (yet) about signal conditioning: https://tinker.koraks.nl/photograph...-the-rgb-cob-led-approach-to-color-enlarging/

 

[bernard_L]

The result: Red had a flat line with no decline over the first second. Green and blue also had flat lines.

Since you have set up the measurement (kudos!) could you report what happens also on longer timescales, say from 1 to 100 seconds. At full power, because that is probably worst-case. Pretty sure that at 1 second, the temperature of the heatsink had no time to rise and is still far from steady-state. Suppose your test strip shows that 5+5+5+5 seconds is perfect; will a continuous 20-second exposure result in the same quantity of light, in the unfavorable case that at 10s continuous, the temperature is rising towards steady-state? (5+5+5+5 is just to keep it simple, actually I'd rather do 5+2+3+4+6+8..... or a similar f-stop progression).
Disclosure: I have no personal experience with LED sources. Some experience in electronics and thermal issues, though.

 

[albada]

I improved the phototransistor circuit, eliminating almost all noise, which let me make two discoveries of interest to anyone building a LED lamp.

Since you have set up the measurement (kudos!) could you report what happens also on longer timescales, say from 1 to 100 seconds.

From a prior posting, the longer times show a drop of red intensity of 0.01 stops.
However, the improved circuit yielded the following start-up trace for green over the first 1.4 seconds:

We see that green at full power declines for about 0.4 seconds after turn-on. I estimate the loss at 0.04 or 0.05 stops. Red is similar. Blue shows no loss at all. Based on the thermal curve for green in the datasheet, I estimate that the junction's temperature rose about 20 degrees C. You might notice a hint more density in a print if your exposure is under one second.

For the record, here is red:

The second discovery is that the Mean Well controller partially averages the PWM signal. At very low duty cycles, the output signal is close to its DC equivalent:

The dashed yellow line at the bottom is for no light. Koraks, the low signal is not being snapped to zero. Rather, it is low-pass filtered, so that the power being output is the same. My reciprocity remark was based on my suspicion that paper might respond differently to pulsed light than to nearly continuous light as graphed above. Manufacturers surely tested with tungsten, but pulsed? We don't know, as you pointed out. Here's a very high duty-cycle that's only 0.1 stop below max:

I talked with an EE at work today about this; he had an idea that makes me wonder if this smoothing was done to reduce EMI (Electro-Magnetic Interference). EMI is an important problem in electronics, and Mean Well would be motivated to reduce it by low-pass filtering their output. I just wish they had described feature this in their datasheet.

Here's the trace for 50% duty cycle:

The absence of sudden transitions (vertical lines) means this will emit low EMI.

I still see some overshoots or ringing periodically at 100 kHz:

Any idea what is causing this? I'm hoping a small capacitor across the Mean Well output-pins will smooth it out of existence.

 

[koraks]

As I said before, to determine the actual behavior of the meanwell supply, scope the actual LED current instead of the phototransistor output. The latter can be subject to all kinds of effects through which rapid transients change shape.

Any idea what is causing this? I'm hoping a small capacitor across the Mean Well output-pins will smooth it out of existence.

That's very well possible and I agree with your colleague it could be there to suppress EMI. Easy enough to figure out; just open up the box and poke around!

 

[albada]

As I said before, to determine the actual behavior of the meanwell supply, scope the actual LED current instead of the phototransistor output. The latter can be subject to all kinds of effects through which rapid transients change shape.

On the Mean Well driver, Out- is not connected to ground, which would force me to scope Out+ and Out- on two scope channels, and mentally subtract the two traces. It makes me wish I had a differential probe. But you're right; it wouldn't hurt to do it anyway to see what's there.

That's very well possible and I agree with your colleague it could be there to suppress EMI. Easy enough to figure out; just open up the box and poke around!

The Mean Well LDD-700L is a 10x23 mm through-hole package that is filled (potted) with epoxy, and epoxy is almost impossible to remove. Sigh.

 

[koraks]

Out- is not connected to ground, which would force me to scope Out+ and Out- on two scope channels

You could just include a series pass resistor of let's say 0.1R and probe across that, no? Or is there a limitation in your scope that prevents this?

(potted) with epoxy

Ahhh, I see. Yeah, that makes it difficult to look into the package. You could try and directly measure capacitance across the output pins; if it's higher than let's say 10nF I'd expect an actual cap to be there. Otherwise you should only measure parasitic capacitances which should be in the pF to single-digit nF range.

 

[bernard_L]

Light output, or paper exposure, versus PWM duty cycle... Does it really matter whether it is linear? That is not something that you will set following the result of a calculation. All you need is that it is monotonic, reasonably smooth, reproducible, and stable versus time (my initial remark). Then you can (assuming b/w) establish a nice map of ISO range versus blue and green PWM duty cycles, or similar for color with which I have no experience. The real tuning variables for exposure are f-stop and time.

As for the thermal drift, as you noted albada, it effect on light output is barely significant; possibly because your design power not very large (can't remember where you mentioned it).

 

[bernard_L]

As I said before, to determine the actual behavior of the meanwell supply, scope the actual LED current instead of the phototransistor output. The latter can be subject to all kinds of effects through which rapid transients change shape.

On the Mean Well driver, Out- is not connected to ground, which would force me to scope Out+ and Out- on two scope channels, and mentally subtract the two traces. It makes me wish I had a differential probe. But you're right; it wouldn't hurt to do it anyway to see what's there.

The industry has anticipated your need; similar components available from several companies. I might even have one for you if interested.

 

[Steve Goldstein]

Does it really matter whether it is linear?

Nonlinearity of light response versus code would make f-stop printing incredibly tedious.

 

[koraks]

The industry has anticipated your need; similar components available from several companies.

A simple series resistor is perfectly adequate for scoping current transients and has unlimited bandwidth. A scope, even a cheap one, has far better signal conditioning than 50 cent sot23 device.
I use the acs712 in my enlarger power supply btw, but not to measure pwm performance.

 

[koraks]

Does it really matter whether it is linear?

It would make filter adjustments in color printing very unintuitive if it isn't. Depending on what the nature of the non-linearity is and when it occurs, there could be additional problems that are even worse to deal with.

Also, using a LUT will be decidedly less nice for color than it is for the 5 or 10 steps you need for B&w. You're easily looking at a 32kbyte table which even by today's embedded standards would be a little peculiar to be content with. It's also quite unnecessary; I just calculate all duty cycles on the fly, both for B&w and color. Frankly if a 3rd order polynomial doesn't cut it, the non-linearity in the hardware is so abysmal that I'd redesign it from the ground up.

 

[koraks]

UPDATE 30 August:

I promised a couple of times to also do a writeup of the actual building project of my color enlarger light source. It took me a while, but I did it. I explained it - all of it - quite expansively in a series of 4 blog posts:

• Part 1: why this project exists anyway, and what it should do
• Part 2: the first generation tests with an RGB COB LED
• Part 3: the second generation and first operational/useful system
• Part 4: the third and current generation color enlarger system

It's a lot of reading...if you're interested but slightly pressed for time, only read parts 1 & 4, or perhaps even only part 4. Parts 2 & 3 are interesting if you wonder about the problems I encountered along the way, which made this project a multi-year (and probably never-ending) exercise instead of something I could have pulled off over a weekend with a little less sleep.

The posts are illustrated with some construction photos, some schematics here and there and lots of explanation of what I did and why. They're also rather idiosyncratic posts, by which I mean they pretty accurately describe the motions I went through, but they are not, and should not be read as, a how-to manual of building a LED color enlarger head. In fact, if you consider anything like this, I think you should read my posts more in the sense of "whatever you do, do not try this because that released the magic smoke in my case and avoid that because it led me to be hospitalized in a mental institution for a while etc."

There are also no scans/photos of prints, test charts and all the other things that would make a writeup like this really sexy, but come on, there's still Pornhub for the visually inclined. I think the limitations of the project are also quite transparently explained, so you can make of this what you will. I 'finished' this project a little over a year ago when the current version went operational, and the early beginnings were in the summer of 2019 (I just found a BOM dated June 2019 of the first generation parts). Over the past 3 years, I tossed a lot of junk (including test strips & prints) and we moved to a new home, so a lot of stuff that might have been 'interesting' is lost forever, or at least pretty much darn lost where I can't find it.

If you want to give a project like this a try, I'd of course be thrilled if you could share your findings with all of us, for example here in this thread, as some of you have been doing already. If I can find the time (which in the short term should be no problem), I'll happily share my thoughts and (unsolicited) advice.

 

[Steve Goldstein]

In part 4, your schematic of the LED current servo has the op-amp inputs interchanged. Non-inverting should be on top, tied to the input; inverting should be on bottom, tied to the FET source and sense resistor.

 

[koraks]

Oh yeah, that's right. Not sure how I missed that. The way it's rigged now would create fireworks. Thanks!

 

[albada]

I promised a couple of times to also do a writeup of the actual building project of my color enlarger light source. It took me a while, but I did it. I explained it - all of it - quite expansively in a series of 4 blog posts:

• Part 1: why this project exists anyway, and what it should do
• Part 2: the first generation tests with an RGB COB LED
• Part 3: the second generation and first operational/useful system
• Part 4: the third and current generation color enlarger system

Thanks for writing these! Actually, I read part 3 last night before you announced it here. I happened to notice it in your blog. Yes, plenty of lessons learned in there.

A question: How well does your current system print colors? Part 4 states that your wavelengths are 660-525-450 for red-green-blue. (For those who don't know, 660 is often called "photo red" and 450 is called "royal blue"). Your article referenced in the original posting left blue-wavelength as an open question, and yet Part 4 says "Color mode really works, and works well." As a result of your experiences and suggestions, I'm planning to make some changes to my LED-head, and I want to double-check on the suitability of those wavelengths for RA4 before buying some.

 

[koraks]

How well does your current system print colors?

Well enough that any limitations I run into are my own instead of the equipment's
At the end of part 4, there's pretty much a direct answer to your question. In short: although I haven't done any color densitometry etc., I can't spot any problems in my color prints. This doesn't mean they aren't there, it's just that I'm not aware of them, even if I look critically. For me, that's good enough at this point.

 

[albada]

For me, that's good enough at this point.

And that's good enough for me as well.

BTW, your article mentioned that your LEDs can consume about 350 watts. Assuming equivalent brightness for tungsten is 6.5x, that's a tungsten-equivalent of 350*6.5 = 2275 watts. Wow! That's close to 10 times the usual 250 watt bulb in enlargers. To improve optical efficiency, have you considered removing the condenser lenses and trays, putting the diffuser about 1 cm above the film, and putting the LEDs 5-10 cm above the diffuser?

 

[koraks]

To improve optical efficiency, have you considered removing the condenser lenses and trays, putting the diffuser about 1 cm above the film, and putting the LEDs 5-10 cm above the diffuser?

Sort of, but haven't got around to it. It's one of those things that seems to have kind of dropped off my to-do list. I planned this at some point, but it didn't materialize.

There's one issue that I'd need to sort out, which is the awkward housing of the light source. This doesn't allow the assembly to be dropped down through the space where the condensors live. Which is just another reason why I need to re-do the physical part of that assembly, as it really was just a quick validation setup that never got changed.
Currently the light source is suspended ca. 25cm (10") above the negative; you're certainly onto something if I drop that.

Btw, in part 4, there's also some discussion (albeit brief) of the diffusor setup. This is kind of problematic, or rather, a balancing act. On the one hand, I could make a more effective diffusor, allowing me to drop the light source much closer to the negative. But the diffusor would eat up a lot of light. The other option, which I ended up doing, is to allow for far more clearance between the light source and the diffusor, allowing me to use a simple sheet of ground glass, which is much more optically efficient.

There's a lot of room for experimentation on this front and your post reminds me that this is something I shouldn't have ignored so long.

 

[albada]

For those interested in building a DIY LED-head, Mean Well has obsoleted their LDD models of LED-drivers, and introduced the new NLDD models. For example, "NLDD-700H" is the 700 mA model. Following koraks' advice to boost my power, I just ordered three NLDD-1400H (the 1.4 amp model) from mouser.com, and I'll soon order some 3 amp Cree XE-G LEDs, which I'll drive at about half-power, leaving lots of headroom.

 

[koraks]

I'm very curious as to how it'll work out, keep us posted! Those LEDs sound very promising.

Concerning the drivers, I'm kind of surprised they only allow up to a 1kHz PWM frequency although the internal switching frequency is 200kHz. Perhaps it has to do with the PWM-wave shaping that we've discussed before. This part still concerns me a bit, I must admit. Might be good for (against) EMI, but not necessarily for printing.

 

[albada]

I'm very curious as to how it'll work out, keep us posted! Those LEDs sound very promising.

Concerning the drivers, I'm kind of surprised they only allow up to a 1kHz PWM frequency although the internal switching frequency is 200kHz. Perhaps it has to do with the PWM-wave shaping that we've discussed before. This part still concerns me a bit, I must admit. Might be good for (against) EMI, but not necessarily for printing.

I'm curious whether the new NLDD models smooth PWM like the prior LDD models. Probably so, as the spec's are similar. They even copied the errors in the datasheet, such as spelling the glue as "expoxy" instead of "epoxy".

Anyway, the scope revealed that at 256 Hz, high and low duty-cycles were smoothed enough to be a rough approximation of DC. So 1000 Hz should smooth much more, perhaps yielding a decent DC-like current. I think that will help printing in two ways: (1) Papers are tested with continuous light, so their response will be what's expected, and (2) moving a dodge/burn tool quickly will produce a smooth blur instead of discreet edges. So I'd like to try quadrupling my PWM to 1024 Hz and see what happens.
Are you aware of a way that such smoothing might hurt printing? If so, I want to know about it so I can watch out for it...

 

[koraks]

Smoothing is likely to be highly non-linear, so you're probably working with an S-curve output for a linear input. You could correct for this of course. And maybe it's not an issue at all.
I'm not concerned about the pulsed light source thing or the burn/dodge thing. Concerning the latter, anything above 20Hz or so won't be a problem.

 

[albada]

Smoothing is likely to be highly non-linear, so you're probably working with an S-curve output for a linear input. You could correct for this of course. And maybe it's not an issue at all.
I'm not concerned about the pulsed light source thing or the burn/dodge thing. Concerning the latter, anything above 20Hz or so won't be a problem.

I use a LUT to convert attenuation (in tenths of stops) to PWM, so some nonlinearity is okay. Presently, the LUT strays from theory by at most 10%, which I'm happy with. Mean Well's datasheet claims it has EMI suppression built-in and that no external caps are needed, so I won't try adding one.

Per Koraks' suggestion, here are the oscilloscope traces of the + and - inputs to the LED-chain at 50% duty-cycle:

I'll guess that LED+ is connected to the sense-resistor on the high side, and that LED- is the drain of the MOSFET. Even when off, there is some voltage between the outputs, as seen on the left end of the scope trace. My multimeter measured 4.something volts, which declined to 1.8V in a few seconds, so I suspect that is residual charge in an EMI-suppression capacitor. We see the rounded corners in the lower trace, which I suppose are exaggerated by the LED's nonlinear V-to-I function. Any comments on these traces? Do they make you want to buy Mean Well LED-drivers?